Independent multi-series forecasting
Three stores with daily sales share a weekly pattern but have different levels, different history lengths (store C opened later) and different exogenous variables (temperature, promotions). ForecasterRecursiveMultiSeries stacks the sliding windows of all stores into a single table with a series column that identifies each store; stores add as many rows as their history allows and missing exogenous variables are filled with NaN. One regressor is fitted on the table and forecasts the next value of every store from its own last values and exogenous variables.
Independent multi-series forecasting
Three stores share a weekly pattern at different levels.
1
Series
2
Training table
3
Fit
4
Predict
Store A
series 0
sales
Store B
series 1
sales
Store C
series 2
sales
skforecast.org
0.3 s